This paper takes a very classic approach to solving the problem of tracking articulated figures. The basic problem is to determine the motion parameters at a sampled time, given the motion parameters from the previous sample. To do this they simply linearize, creating the classical Jacobian formulation, AJx = B, where x is the set of motion parameters, J is their Jacobian, A is the spatial gradient of the image, and B is the temporal gradient of the image.
The novel work in the paper is the integration of multiple views to aid in determining the correct 3D motion of the tracked object. Since single camera views are inherently ambiguous, using multiple views allows many of these abiguities to be resolved correctly.
Because the formulation is linear, adding more views is simply a matter of extending the system. By selecting views in orthogonal directions, motion degeneracy - such as motion along the optical axis of a camera - can be resolved.
This paper takes essentially the same approach as the previous paper in solving the human body tracking problem. Both approaches require an initial placement of the model, and both use image gradients, along with the Jacobian of the parameter space, to formulate a linear solution model.
The key distinction between this work and the other paper is not clear, they both model all six degrees of freedom at each joint, and they both make use of multiple camera views. However, the description in each paper is very different. In the Yamamoto paper, the algoithm is laid out in a very straightforward manner with little theoretical discussion. The experiments are implemented in a practial manner as well. In the Bregeler paper the description is more general and more theoretical. The experiments used to describe the results are also based on recorded image sequences from the turn of the century.
The one thing that does set the two methods apart is Bregeler's use of exponential maps in modeling the kinematic chain. Unfortunatly, no evidence is given that doing this helps the tracking process.